Prompt Engineering

Livro digital

Título:
Prompt Engineering

Autor:
Lee Boonstra

Categoria:
Tecnologia > IA

Doador:
Raffaello D. N.

Sinopse:
Writing a prompt feels easy until you need precision. This Google whitepaper treats prompt engineering as a systematic discipline, walking through output configuration—temperature, top-K, top-P—then advancing through a taxonomy of techniques: zero-shot baselines, few-shot calibration, role and contextual prompting, and the reasoning family of Chain of Thought, self-consistency, Tree of Thoughts, and ReAct. The second half tackles production concerns: code generation and debugging patterns, schema-enforced JSON output with automatic repair, and the iterative workflow of documenting prompt variants. Practical chapters on multimodal prompting and model-specific optimization close the gap between research techniques and what works reliably in Vertex AI and the Gemini API. Concise and vendor-aware without being a sales pitch, this whitepaper doubles as a field manual for developers who need LLM outputs that are not just plausible but correct, structured, and reproducible.